Last week at SuperComputing Conference 2017, AMAX released the [SMART]Rack P47, a PetaFLOP-In-A-Rack solution featuring AMD CPUs and GPUs, designed for deep learning and rendering applications.
The [SMART]Rack P47 combines the power of 20x AMD EPYC™ Processors and 80x AMD Radeon Instinct™ GPUs, based on AMD’s revolutionary “Vega” architecture, to deliver one PetaFLOP of 32-bit FP single precision compute power at a stunning 30 GigaFLOPs/Watt. It is designed to address today’s most complex computational problems in Artificial Intelligence, Deep Learning, VDI, advanced rendering, compute and research at an unmatched performance-per-dollar and performance-per-watt level.
The [SMART]Rack P47 PetaFLOP-Performance-in-a-Rack solution made its first appearance at SIGGRAPH 2017 in July, when AMD CEO Lisa Su proudly unveiled the rack and described it as “the most beautiful server rack that has ever been built.”
The supercomputing-class performance of the fully-integrated rack is achieved by 20x ServMax™ P47 servers, designed with high scalability to support a variety of workloads such as virtual desktop infrastructure, deep learning, machine learning, advanced rendering and compute. Each ServMax™ P47 server supports 1x AMD EPYC™ CPU and 4x AMD Radeon Instinct™ MI25 PCIe GPUs, delivering 20,480 Stream Processors, 49.20 TFLOPS single-precision, and 98.40 TFLOPS half-precision performance. Integrated with AMD’s ROCm open software platform and MIOpen libraries, the ServMax™ P47 servers deliver superior performance with unmatched manageability for deep learning training deployments in datacenters.
The [SMART]Rack P47 solution also integrates advanced memory and storage technologies; unified 100Gb InfiniBand/ Ethernet HPC fabric for increased in-rack bandwidth and productivity; integrated rack level out-of-band management layer for remote server health monitoring and rack orchestration; and a high-density rack cooling system add-on for optimal efficiency in heat exchange.
“As a high-performance technology provider enabling enterprises to close the gap between scale up performance, compute density and cost, AMAX sees the [SMART]Rack P47 as a game changer,” said Julia Shih, VP of Business Development, AMAX. “Starting from a single ServMax™ P47 server, we can scale upwards to supercomputing-class performance by leveraging AMD EPYC CPUs, AMD Radeon Instinct, and AMD ROCm development tools to support Deep Learning, rendering, and a host of other applications.”
AMAX is now taking pre-orders for both the ServMax™ P47 server and the fully-integrated [SMART]Rack P47, with delivery estimated in Q1 of 2018.